Uncertainty in latent representations of variational autoencoders optimized for visual tasks

2024-04-23

arXiv.org

AI Summary

Standard VAEs often suffer from distorted uncertainty representations in latent variables, leading to unreliable uncertainty estimation in vision tasks such as image inpainting, latent-space interpolation, and out-of-distribution (OOD) detection. To address this, we propose the Explaining-Away VAE (EA-VAE), the first VAE framework to integrate the classical computer vision principle of “explaining away” with divisive normalization—a biologically inspired neural computation paradigm—into the variational inference process. EA-VAE introduces a global latent variable to explicitly model competitive explanatory relationships among local features, thereby enforcing posterior uncertainty that adheres to Bayesian normative principles. Experiments demonstrate that EA-VAE significantly improves the fidelity and calibration of uncertainty quantification across image restoration under degradation, smooth latent interpolation, and OOD detection. It consistently outperforms standard VAEs and state-of-the-art variants in both uncertainty reliability and downstream task performance.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Interpretability, Explainability, and Transparency
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Deep Generative Models (DGMs) can learn flexible latent variable representations of images while avoiding intractable computations, common in Bayesian inference. However, investigating the properties of inference in Variational Autoencoders (VAEs), a major class of DGMs, reveals severe problems in their uncertainty representations. Here we draw inspiration from classical computer vision to introduce an inductive bias into the VAE by incorporating a global explaining-away latent variable, which remedies defective inference in VAEs. Unlike standard VAEs, the Explaing-Away VAE (EA-VAE) provides uncertainty estimates that align with normative requirements across a wide spectrum of perceptual tasks, including image corruption, interpolation, and out-of-distribution detection. We find that restored inference capabilities are delivered by developing a motif in the inference network (the encoder) which is widespread in biological neural networks: divisive normalization. Our results establish EA-VAEs as reliable tools to perform inference under deep generative models with appropriate estimates of uncertainty.

Problem

• Variational Autoencoders
• Uncertainty Estimation
• Image Anomaly Detection

Innovation

• EA-VAE
• Uncertainty Estimation
• Normalization Pattern